Most observability failures are silent: the call succeeds, but the log, metric or event never shows up where you look. Check the silent-failure tables first, then the exceptions.

Silent failures

Symptom	Cause → fix	Check
A custom metric isn't in the console or `list-metrics`	A new metric takes up to 15 minutes to appear in `ListMetrics`. Back-dated data (more than 3 hours old) can take hours to become queryable. Check the namespace and Region	`aws cloudwatch list-metrics --namespace <ns> --metric-name <m>` after 15 minutes
The metric exists but the graph or alarm is empty (`INSUFFICIENT_DATA`)	The dimensions you query must match exactly what you published, unit included. Custom metrics aren't aggregated across dimensions	`list-metrics` output: compare each `Dimensions` entry with your query, exactly
EMF metrics never appear	Check the `AWS/Logs` namespace for extraction failures. Likely causes: the event isn't purely the JSON object; dimension keys or values aren't top-level; a value is out of range; an array has more than 100 items; more than 100 metric definitions; a timestamp outside 14 days past to 2 hours future (`reference/cloudwatch/metrics-and-math.md`)	`aws cloudwatch list-metrics --namespace AWS/Logs` for EMF error metrics; read a raw event in the log group
Some log events missing after `PutLogEvents` succeeded	Events more than 2 hours in the future or older than 14 days or the retention period are rejected individually. Read `rejectedLogEventsInfo` in the response	The `PutLogEvents` response → `rejectedLogEventsInfo` (`tooNewLogEventStartIndex`, `tooOldLogEventEndIndex`, `expiredLogEventEndIndex`)
A log group has no logs from Lambda	The function never ran, or its role can't write logs (`guides/lambda/debug-failing-function.md`). Or it logs to a custom group set in `LoggingConfig`	`aws lambda get-function-configuration --function-name <fn> --query LoggingConfig` → `LogGroup`; the function's `Invocations` metric
`console.debug` lines missing from Lambda	JSON log format defaults the application log level to INFO (`reference/cloudwatch/log-groups.md`)	Same call → `LogFormat`, `ApplicationLogLevel`
Old logs "disappeared"	Retention expired them. Deletion runs up to 72 hours after expiry	`aws logs describe-log-groups --log-group-name-prefix <g>` → `retentionInDays`
Logs in a KMS-encrypted group can't be read	The key is disabled, the logs principal was removed from the key policy, or the key was deleted, which makes the data unrecoverable. A newly associated key takes up to 5 minutes to apply	`describe-log-groups` → `kmsKeyId`, then `aws kms describe-key --key-id <k> --query KeyMetadata.KeyState`
Alarm fired once and never notified again	Actions run on state change only (`guides/cloudwatch/create-alarms.md`)	`aws cloudwatch describe-alarm-history --alarm-name <a> --history-item-type StateUpdate`

CloudWatch Logs exceptions

Exception	Cause → fix	Check
`ResourceNotFoundException`	The log group or stream doesn't exist in this account and Region. Create it, or check the name (`/aws/lambda/<fn>`) and Region	`aws logs describe-log-groups --log-group-name-prefix <g> --region <r>`
`ResourceAlreadyExistsException`	Creating a group or stream that exists. Treat it as success in setup code	Same call
`InvalidParameterException`	A malformed request. For `PutLogEvents`, commonly events not in chronological order, a batch spanning more than 24 hours, more than 10,000 events, or over 1,048,576 bytes (messages plus 26 bytes per event). For queries: a bad log group identifier, such as an ARN with a trailing `*`	The message names the field; sort events by timestamp and check the batch's time span
`InvalidSequenceTokenException`, `DataAlreadyAcceptedException`	No longer returned, because `PutLogEvents` ignores sequence tokens. Delete the retry logic for them	Search your code for `sequenceToken`
`MalformedQueryException`	Logs Insights syntax error. The `QueryCompileError` gives the location	The error's `QueryCompileError` → `location`
`LimitExceededException`	A maximum was reached, for example concurrent Logs Insights queries (shared with dashboards and scheduled queries). Back off	`aws logs describe-queries --status Running`
`ThrottlingException`	The per-account TPS quota for that API was exceeded (`PutLogEvents`, `Create`, `Describe`: adjustable in Service Quotas). The SDK retries it; batch more per call	Calls per second for that API (SDK retry logs, CloudTrail `errorCode`)
`ServiceQuotaExceededException`	A resource quota, such as the number of log groups. Request an increase	Service Quotas for CloudWatch Logs
`AccessDeniedException`	An IAM denial (`reference/iam/errors.md`). On `associate-kms-key` / `create-log-group --kms-key-id`, a missing `kms:DescribeKey` on the key	`aws sts get-caller-identity`; for KMS, `aws kms get-key-policy` for `logs.<region>.amazonaws.com`
`OperationAbortedException`, `ConflictException`	A concurrent change to the same resource. Retry	Another process changing the same group (CloudTrail)
`UnrecognizedClientException`	Invalid access key ID or secret key (`reference/sts/credential-sources.md`)	`aws sts get-caller-identity` with the same credentials
`ServiceUnavailableException` (500)	Transient. Retry with backoff	Retry succeeds; the AWS Health Dashboard
`SessionStreamingException`, `SessionTimeoutException`	Live Tail session errors. Restart the session	Restart `StartLiveTail` and see if it recurs

CloudWatch metrics and alarms exceptions

The PutMetricData causes are documented. The rest are the hub's reading of the SDK exception names.

Exception (SDK name; API error code)	Cause → fix	Check
`InvalidParameterValueException` (`InvalidParameterValue`)	A value out of range or invalid: NaN or Infinity, beyond ±2^360, a timestamp more than 2 weeks old or 2 hours ahead, a bad namespace, or more than 30 dimensions	Print the values you send: NaN, Infinity, out-of-range or the timestamp
`InvalidParameterCombinationException` (`InvalidParameterCombination`)	Parameters that can't go together. On alarms, for example, both `MetricName` and `Metrics` math (UNVERIFIED example)	The request JSON: which mutually exclusive fields are both set
`MissingRequiredParameterException` (`MissingParameter`)	A required field is absent	The message names the field
`LimitExceededException` / `LimitExceededFault`	A quota such as the number of alarms. Request an increase or consolidate	Service Quotas for CloudWatch; `aws cloudwatch describe-alarms --query 'length(MetricAlarms)'`
`ResourceNotFoundException`, `DashboardNotFoundError`	Wrong alarm, dashboard or other resource name, or the wrong Region	`aws cloudwatch describe-alarms --alarm-names <a> --region <r>` or `list-dashboards`
`DashboardInvalidInputError`, `InvalidFormatFault`	Dashboard body or input JSON is invalid	The error's `dashboardValidationMessages`
`KmsAccessDeniedException`, `KmsKeyDisabledException`, `KmsKeyNotFoundException`	A KMS key used by a CloudWatch feature can't be used (`reference/kms/key-policies-and-errors.md`)	`aws kms describe-key` and `get-key-policy` for the key the feature uses
`ConcurrentModificationException`	A parallel update. Retry	Another deploy updating the same resource (CloudTrail)
`InternalServiceFault` (`InternalServiceError`, 500)	Transient. Retry	Retry succeeds

When none of these fits

Check where the data should be, step by step: did the producer call succeed (CloudTrail PutLogEvents / PutMetricData / PutEvents with errorCode, guides/cloudtrail/find-who-did-what.md)? Is it the right account, Region, log group, namespace and dimensions? Is it within retention, and past the visibility delays above? Collect the request ID from the SDK error for AWS Support.